Neural State Monitoring in the Treatment of Epilepsy: Seizure Prediction—Conceptualization to First-In-Man Study

Abstract

This research study is part of a therapy development effort in which a novel approach was taken to develop an implantable electroencephalographic (EEG) based brain monitoring and seizure prediction system. Previous attempts to predict seizures by other groups had not been demonstrated to be statistically more successful than chance. The primary clinical findings from this group were published in a clinical paper; however much of the fundamental technology, including the strategy and techniques behind the development of the seizure advisory system have not been published. Development of this technology comprised several steps: a vast high quality database of EEG recordings was assembled, a structured approach to algorithm development was undertaken, an implantable 16-channel subdural neural monitoring and seizure advisory system was designed and built, preclinical studies were conducted in a canine model, and a First-In-Man study involving implantation of 15 patients followed for two years was conducted to evaluate the algorithm. The algorithm was successfully trained to correctly provide a) notification of a high likelihood of seizure in 11 of 14 patients, and b) notification of a low likelihood of seizure in 5 of 14 patients (NCT01043406). Continuous neural state monitoring shows promise for applications in seizure prediction and likelihood estimation, and insights for further research and development are drawn.

Deep Convolutional Neural Network Based Interictal-Preictal Electroencephalography Prediction: Application to Focal Cortical Dysplasia Type-II

“…Importantly, the majority of previous deep learning-based studies focused on either all intact electrodes ( 5 , 6 , 35 ), a fraction of all electrodes chosen by a clinical examination ( 10 ), or an entropy-based algorithm ( 9 ) that was selected in advance of model training. Furthermore, recent seizure advisory systems typically consist of a 16-electrode implantable device based on multiple 4-electrode strips that collect long-term human intracranial EEG recordings for conventional machine learning ( 31 , 32 ) and deep learning ( 35 ) techniques. Finally, cortical stimulation devices designed to detect seizure occurrences are typically made up of 8 or 16 electrodes with multiple 4-electrode strips ( 29 , 30 ).…”

“…However, they reported highly varied distributions of the HFOs among patients and poor performance outcomes in their seizure predictors. Recent studies using implantable devices for seizure detection and prediction acquired intracranial EEG recordings at sampling frequencies lower than 400 Hz in human ( 29 – 32 , 35 ) and canine ( 37 ) subjects, likely to lower power consumption and increase data processing efficiency. Our results suggest that sampling frequencies from 128 to 512 Hz have no significant impact on interictal-preictal discriminability by CNN-based classification models.…”

“…Four, eight, and 16 electrodes were selected according to their clinical significance ranking in each patient. Since recent seizure detection (29) and prediction (32) devices have used multiple four-electrode strips, the minimum number of electrodes was set to four.…”

“…Patients with FCDs undergo presurgical evaluations using electrophysiology and functional imaging to identify the extent of their epileptogenic zones. Those who suffer from medically refractory seizures are, in some cases, even encouraged to use implantable devices for seizure detection ( 29 , 30 ) and prediction ( 31 , 32 ). These implanted detection systems contain enough computational power to allow for long-term EEG monitoring while still assuring the patient's comfort, thereby allowing for timely medical interventions in a patient-specific manner.…”

Neurovista: Concept to first-in-man: The war story behind launching a venture to treat epilepsy

“…These feature extractors were then vetted against the EEG data and narrowed down from many thousands to 288, from which a customized subset of a handful of feature extractors was selected for each patient. The development of the technology is described in more detail in DiLorenzo 2019,[2] and the pilot clinical study was described in Cook 2013. [1] Ultimately, the company went on to raise $71.2 million in predominantly venture financing with a small amount of strategic investment over an 11- year timeline following founding in 2002.…”